Wind Power Gearbox Bearing Fault Detection Approach Based On Information Fusion

With the shortage of fossil energy and the deterioration of natural environment,countries around the world pay increasing attention to the development and utilization of wind energy,renewable and clean energy.As an important mechanical equipment to convert wind energy into electric energy,the rush to instal l wind turbine is becoming increasingly obvious.However,with the large size and the exponential speed of new installations,wind turbines,which have been under harsh working conditions for a long time,have gradually entered the stage of high incidence of failures and frequent accidents.Among the high annual maintenance costs,gearboxes are among the components that have the longest downtime and highest repair costs after a failure.Further research found that 76% of gearbox failures are caused by bearing failures.Therefore,timely detection and discovery of faults in gearbox bearing operation can effectively reduce gearbox failure rate and thus improve wind farm efficiency.In order to improve the fault detection accuracy of wind turbine gearbox bearing,the corresponding research is carried out from the signal level and feature level of information fusion.The specific research contents are as follows.(1)A signal-level fusion based on least squares support vector machine algorithm is proposed for wind turbine gearbox bearing fault detection.Firstly,the correlation degree between the whole year data of multiple wind turbines and the bearings is analyzed by distance correlation algorithm,and the four sets of original input parameters most relevant to the bearings are selected comprehensively.Then the signal-level fusion of the four sets of original input parameters is performed by an extended Kalman filter algorithm.Finally,the fused values are used as the sample set of least squares support vector machine algorithm for fault detection.The testing results show that the average detecti on accuracy of the fault detection model after signal-level fusion is improved by 0.002 and the average running time is shortened by0.210 s,which fully proves the effectiveness of the fault detection model based on signal-level fusion.(2)A wind turbine gearbox bearing fault detection method based on an isolated forest algorithm with feature-level fusion is presented.Firstly,a modified algorithm of distance correlation algorithm-multiscale graph correlation algorithm is introduced into the fault detection field as a nonlinear correlation algorithm for selecting original input parameters,and nine groups of original input parameters most relevant to gearbox bearings are selected.Secondly,two types of feature vectors,root mean square and envelope of original input parameters are extracted respectively.Then a self-organizing mapping neural network is used.Finally,the fused values are used as training samples for fault detection of wind turbine gearbox bearings using the isolated forest algorithm.Compared with the other three commonly used intelligent algorithms,fault detection based on the isolation forest algorithm has the best detection performance.Further compared with the pre-feature-level fusion,the average detection accuracy is improved by 0.2 07 and the average running time is shortened by 0.072 s.The above study shows that t he fault detection model based on feature-level fusion can effectively shorten the detection time while accurately identifying faults.